SuperValid: Capability-Aligned OOD Validation for Generalizable Downstream Scaling 事件

SuperValid: Capability-Aligned OOD Validation for Generalizable Downstream Scaling arXiv:2605.28179v1 Announce Type: new Abstract: Scaling laws guide large language model training by relating compute to cross-entropy loss, and recent work further extends them to predict downstream benchmark performance. However, prior approaches face generalization limitations from two aspects: focusing on benchmark-level performance introduces scenario-specific artifacts, while relying on IID validation loss f